Just a bit more "data" in the form of images. I started with a "random polyhedron" built from the convex hull of a small number of points. Then I iterated the process of replacing each face by its centroid, i.e., $\alpha=\beta=\gamma=\frac{1}{3}$ in my original description, and again taking the convex hull. At least visually, the data support Gjergji's suggestion that the process converges to ellipsoids (of different axes lengths). <br /> ![CentroidPolyhedra][1] <br /> So, to make an explicit conjecture out of these observations: > **Conjecture**. Given any convex polyhedron $P$ in $\mathbb{R}^3$, let $c(P)$ be the convex hull of the centroids of the faces of $P$. Then $c^k(P)$ converges to an ellipsoid as $k \to \infty$. Secondarily, it makes sense to conjecture the same holds for any convex polytope $P$ in $\mathbb{R}^d$. [1]: https://i.sstatic.net/mGkVw.jpg